Extensions/ComfyUI Krea 2 NegPiP
ComfyUI Extension

ComfyUI Krea 2 NegPiP

NegPiP-style negative prompt weighting for Krea 2

By Brandonn-Etheve·Created 2 days ago·Updated a day ago· 0
Brandonn-Etheve/ComfyUI-krea2-negpip
Nodes—
On cloudLocal install
Stars0
Updateda day ago
Readme

ComfyUI Krea 2 NegPiP

ComfyUI Krea 2 NegPiP screenshot

ComfyUI custom node that adds NegPiP-style negative prompt weighting support for Krea 2 workflows.

Use prompt weights such as (word:-1.2) to suppress specific concepts in Krea 2 prompts.

Installation

Clone or copy this repository into your ComfyUI custom_nodes directory:

cd ComfyUI/custom_nodes
git clone https://github.com/blue-pen5805/ComfyUI-krea2-negpip.git

Restart ComfyUI after installation.

Usage

  1. Load a Krea 2 model and Krea 2 CLIP as usual.
  2. Add the Apply Krea2 NegPiP node.
  3. Connect both the MODEL and CLIP through this node before text encoding and sampling.
  4. Use negative prompt weights in the prompt, for example:
a portrait photo, (blurry:-1.0), (low quality:-1.2)

The node returns a patched MODEL and CLIP. Use those outputs for the rest of the workflow.

Inputs

  • model: Krea 2 model.
  • clip: Krea 2 CLIP loaded with CLIPLoader type krea2.
  • value_strength: Strength of the negative prompt effect. Default: 1.0.
  • patch_txtfusion_refiners: Optional stronger effect path. Default: false.
  • block_start: First transformer block to affect. Default: 0.
  • block_end: Last transformer block to affect. Default: 27.
  • block_stride: Affect every Nth block in the selected range. Default: 1.
  • debug: Optional diagnostics: off (default), log, or measure. measure runs an additional model forward pass for comparison.

Notes

  • This node is intended for Krea 2 model layouts only.
  • Negative weights are parsed from ComfyUI prompt weighting syntax, for example (token:-1.0).
  • Positive and non-unit weights are still applied to the text conditioning magnitude.
  • After updating this node, restart ComfyUI and re-run text encoding.
  • Prompts with image placeholders or custom embeddings are supported conservatively. Negative weights after image embeddings may not always be applied.

Credits

This project is inspired by and references the original NegPiP implementation: